Evaluation of different digital elevation models for analyzing drainage morphometric parameters in a mountainous terrain: a case study of the Supin–Upper Tons Basin, Indian Himalayas

Topography is a key controlling factor in the operation of a variety of natural processes (Summerfield and Hulton 1994; Montgomery and Brandon 2002). Hence it needs to be quantitatively analyzed (Pike 2000; Lague et al. 2003), to ascertain the relative efficacy of its constituents and operative mechanisms (Brierley et al. 2006; Phillips 2007), and to gauge the response of geomorphic systems to different stimuli (Phillips 2006, 2009; Ahmed et al. 2010). Rivers are one of the most sensitive elements of the landscape (Brunsden 2001; Thomas 2001; Smedberg et al. 2009), and fluvial systems represent a long-term adjustment of streams (Whipple 2001; Tucker 2004), to the varying conditions of climate, lithology and tectonics (Burt 2001; Kirby and Whipple 2012; Whittaker 2012). Changes in the prevailing climatic conditions (Bogaart and van Balen 2000; Huisink 2000; Wobus et al. 2010), base levels (Blum and Tornqvist 2000; Stokes et al. 2002), and/or tectonic situations (Whipple 2004), may trigger short and long term responses by fluvial systems in the form of channel morphological adjustments (Rinaldi 2003), discharge and sediment regime changes (Whipple and Tucker 2002), and re-sculpting of the riparian landforms and landscape (Vandenberghe 2002; Nicholas and Quine 2007; Rittenour et al. 2007). These responses, particularly to structural disturbances and tectonic forcing, are usually manifested in the form of major anomalies in the morphometric attributes of rivers and their drainage network (van Heijst and Postma 2001; Church 2002; Lin and Oguchi 2006; Thomas et al. 2010, 2012; Bali et al. 2011; Bahrami 2013). Although recent researches have focused more on examining processes, materials and chronology (e.g. Lewin et al. 2005; Chiverrell et al. 2009; Hooke 2008; Trimble 2009; Solleiro-Rebolledo et al. 2011), the systematic evaluation of land surfaces and drainage characteristics remains a central theme in geomorphology (e.g. Cammeraat 2002; Minar and Evans 2008; Siart et al. 2009; Paik and Kumar 2010; Prasannakumar et al. 2013). Consequently, geomorphometry (i.e., the science of the measurement of landforms), occupies an important domain in the discipline (Rao 2002; Wobus et al. 2006; Bishop et al. 2012; Evans 2012).

This ‘geomorphometry’ may be classified into two types—‘general geomorphometry’, which analyses the overall land surface form, and ‘specific geomorphometry’, which examines the characteristics of individual landforms (Evans 2012). Widespread application of general geomorphometry, particularly in drainage basin analysis can be observed (e.g. Vorosmarty et al. 2000; Jordan et al. 2005; Lindsay 2005; Wood 2009; Hayakawa and Oguchi 2009; Cavalli et al. 2013). These morphometric properties of a drainage basin are the quantitative attributes of the landscape, derived from the terrain, the elevation surface and the drainage network (Goudie 2004), and include size, relief, surface, shape and texture attributes. Their calculation is the first step in geomorphometry and quantitative geomorphology. Evaluation of these parameters also provides a basis for ascertaining the structural and lithological controls inherent in the landscape, as well as understanding the tectonic history of the river basins under consideration (Ferraris et al. 2012; Jacques et al. 2014).

Digital elevation models (DEMs) have been frequently used for the above morphometric analysis of river basins through the extraction of topographic parameters and stream networks, and their use presents many advantages over traditional topographical maps. A DEM may be defined as a regular gridded matrix representation of the continuous variation of relief over space (Burrough 1986), and is a digital model of the land surface form. The primary requirement of any DEM is that it should have the desired accuracy and resolution and be bereft of data voids (Sefercik and Alkan 2009). Their steady and widespread application can be further attributed to their easy integration within a GIS environment (Moore et al. 1991; Weibel and Heller 1991). Before the year 2000, the base elevation models depicting a global coverage were available in a 1 km resolution like GTOPO-30 (Global Topography in 30 arc-sec) and GLOBE (The Global Land 1 km- Base Elevation Project) (Sefercik and Alkan 2009). However, in the last decade, more advanced global DEMs with better resolutions have become available, like the Shuttle Radar Topography Mission (SRTM) (version 4, C-Band DEM of 3 arc-second, 90 m resolution) and the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) (version 2, 30 m resolution) DEMs which have mitigated the problem of spatial resolution to a great extent. For users within India, or those seeking to examine Indian landscapes, the available DEM dataset library received another member through the release of the CartoDEM data (version 1, only for Indian territories) at 30 m in 2011. Apart from these freely available ready-made DEM datasets, purchased stereo-images from a number of satellites (e.g. Cartosat 1, Landsat 7 ETM+, QuickBird, IKONOS, SPOT, ASTER sensors, among others) have also been used to create DEMs using various software applications for examining landscapes (Toutin et al. 2001; Toutin 2002, 2004; Poli et al. 2002; Hirano et al. 2003; d’Angelo et al. 2008; Deilami and Hashim 2011; Giribabu et al. 2013).

The biggest advantage of DEMs over traditional topographical maps is the seamless provision of data having a global coverage. Due to their wide applicability and ease of use, DEMs have been used in a variety of studies where terrain and drainage factors play prominent roles. Numerous studies on morphometric analysis from DEMs have been carried out across the world in recent years (e.g. Dietrich et al. 1993; Nag 1998; Snyder et al. 2000; Lindsay et al. 2004; Korup et al. 2005; Mesa 2006; Deng 2007; Ehsani and Quiel 2008; Lindsay and Evans 2008; Wilson et al. 2008; Wang et al. 2010; Ferraris et al. 2012; Caraballo-Arias et al. 2014; Jacques et al. 2014). In India, some prominent studies where DEMs have been employed for river basin analysis, estimation of soil loss, water resource evaluation and topographic characterization include Chopra et al. (2005), Kale and Shejwalkar (2007), Rudraiah et al. (2008), Sreedevi et al. (2009), Patel and Sarkar (2010), Malik et al. (2011), Pareta and Pareta (2011), Agarwal et al. (2012), Patel et al. (2012), Altaf et al. (2013), Magesh et al. (2012, 2013), Agarwal et al. (2013), Dar et al. (2013), Prabu and Baskaran (2013), Singh et al. (2013, 2014), Aher et al. (2014), Ambili and Narayana (2014), Magesh and Chandrasekar (2014), and Ghosh et al. (2015), among others. DEM usage in drainage routing and flood prediction too has gained popularity (e.g. Ozdemir and Bird 2009; Youssef et al. 2011; Sreedevi et al. 2013)

Normally it is accepted that higher resolution DEMs are more precise (Saran et al. 2009), and that this higher precision implies a greater degree of accuracy and a finer extraction of the land surface components, especially slope facets (Dragut and Blaschke 2006), and drainage lines (Anornu et al. 2012; Srivastava and Mondal 2012). Hence, the search for an optimal cell resolution and cell-size of DEMs has been a topic of research in the last few years (Hancock et al. 2006; Sharma et al. 2009; Sreedevi et al. 2009; Ahmed et al. 2010). In such geomorphometric analysis, the DEM resolution governs the scale of the features extracted (Hengl and Evans 2009), with the morphometric attributes extracted being also scale dependant (Dragut et al. 2009). A number of studies have delved into the accuracy assessment of individual DEM datasets, e.g. for SRTM data (Gorokhovich and Voustianiouk 2006; Weydahl et al. 2007), or for ASTER data (Eckert et al. 2005; San and Suzen 2005; Cook et al. 2012), and have looked into their effect on the extracted features like drainage (Fujita et al. 2008; Li and Wong 2010; Tarekegn et al. 2010) and terrain aspects (Zhou and Liu 2004; Vaze et al. 2010). Previous studies have also shown that the pre-release ASTER-GDEM had yielded better results than the SRTM-DEM in western Japan (Hayakawa et al. 2008); but its post-release version was found to be inferior to the SRTM-DEM in the mountainous regions of Turkey (Sefercik 2012). However, instances where two or more sets of DEMs were compared with respect to their morphometric parameters (e.g. Lindsay and Evans 2008; Taramelli et al. 2008; Hirt et al. 2010; Hosseinzadeh 2011; Suwandana et al. 2012; Mukherjee et al. 2013; Gopinath et al. 2014), are relatively few, especially in mountain landscapes of India. Furthermore, often in this accuracy assessment, the focus is more on comparison of absolute elevation parameters with lesser focus given to investigating how the various morphometric variables that are derived, vary from one dataset to the other, as well as how their prepared maps differ. Therefore, in this study, the morphometric properties of the Supin–Upper Tons watershed, located amidst the Garhwal Himalayas in India, are initially computed from different DEMs as well as topographical maps, and then subsequently mapped and compared in order to ascertain the most reliable source of digital elevation data for geomorphometry, generally overall and particularly for such regions.

This study compares elevation profiles, stream networks and morphometric parameters derived from freely available DEM products as well as from topographical maps of different scales for a chosen mountainous river basin. The comparison of these aspects, checked against field collected elevations at select Ground Control Points (GCPs) via a GPS survey, helped to ascertain which of the data products are more consistently able to represent the actual topographic features and are most useful in extensive drainage line demarcation, computation of stream statistics and enumeration of allied morphometric parameters. The study also highlights the extent of map scale or DEM resolution on terrain and drainage parameter extraction and how their respective maps differ as a result. To nullify this scale-effect, the different preliminary products were also re-sampled to a common resolution for better comparison and analysis, with the results tabulated.

The Supin–Upper Tons Basin comprises part of the Tons River (the largest tributary of the Yamuna River) Basin in the Purola Tehsil of the Garhwal Region of Uttarakhand state, India, located between 78°06′E–78°38′E and 31°00′N–31°17′N. The River Supin is one of the principal tributaries of the Tons River, which itself is a tributary of the Yamuna River (Pankaj et al. 2012). The other major tributary, the Obra Gad, merges with the Supin River near the village of Fitari. The Tons River is initially formed by the joining of its tributaries, the Har-ki-dun Gad and the Ruinsara Gad, before their combined flow meets the Supin River (Fig. 1). The basin covers an area of about 977 sq km approximately, having a perimeter of about 180 km. The Supin River itself originates from the snout of the Khimloga Glacier while the main stream, the Tons River, emerges from the Banderpunch Glacier and these rivers converge near Sankri village, about 30 km downstream from their respective sources. The basin mouth is near Netwar village where the combined flow of the Supin–Tons merges with the Rupin River.

The physiography of the watershed is dominated by high mountain ranges and steep spurs alternating with deep declivities, i.e. a landscape of sharp divides and entrenched river courses. The altitude ranges from 1200 m to 6387 m (Krishan et al. 2009). A substantial part of the basin is over 4000 m elevation above mean sea level. The highest point in the basin is the Banderpunch Peak (6315 m). The areas above 3000 m are more or less glaciated. Forests, agricultural tracts, snow covered hillslopes, glaciers and grasslands are the major land cover and land use types.

The rainfall received over the basin area varies from 1000 to 1500 mm annually, with occasional heavy cloud bursts and the area is subjected to regular snowfall, with significant amounts occurring between October to May (Krishan et al. 2009). Approximately 49 % of the entire basin area, especially its upper reaches, is under perennial snow cover (as ascertained from the IRS-P6 LISS-III image of the study area of October 2008, obtained from the Bhuvan Portal, after digitisation of the visible snow cover extent, and also subsequently verified from the USAMS and SoI topographical sheets). Past glacial retreat may be inferred from the present ‘U’ shaped valleys with moraines and aggradational slopes present, downstream from the present glacial snouts.

Geologically, the rocks exposed within the basin range from the Proterozoic to the Cambrian in sequence and age. The younger rocks occupy the northern and eastern parts while the older Proterozoics are divided into a number of tectonic groups in the west and south (GSI 2004). The area is cut across by a number of thrusts, namely the Purola Thrust, Main Central Thrust and Jutogh Thrust, which indicate the dynamic pressures the rocks have been subjected to and account for the large varieties of metamorphics seen here. Near the basin mouth, the Jaunsar Group comprises of rocks of Neo Proterozioc age with constituents like grey and green phyllites, quartzites and schists. The Purola Crystalline Group thrusts over the Jaunsar Group (via the Purola Thrust), in the lower basin portion, and contains amphibolite, pebbly conglomerate gneiss, biotite schists and quartz. The Central Crystalline Group, occupying the middle part of the basin, can be divided into lower grade and higher grade categories. The lower grades (called the Gangar Formation), comprise of inter-calated sequences of schists, mica, quartzites, biotites, quartz and gneiss. The higher grades (called the Har-ki-dun Formation) thrust over these lower grades (via the Main Central Thrust), and comprise of schist, gneiss, migmatites and basic intrusives. Emplacement of biotite granite (Rakcham Granite of Palaeozioc age) has also occurred in the central part of the basin. The eastern and northern portions comprise of the relatively younger Haimanta Division of the Early Cambrian rocks, which is further divisible into the Batal Formation and the Kunjanla Formation. The main rock types in these formations are grey phyllite, quartzite, carbonaceous shale and green shale. Intrusives of metamorphosed granite of the Paleozoic Era are also present in this area. The general dip of the rocks is NW–SE (Pankaj et al. 2012), and these have been subjected to intense deformation in the form of folding, thrusting and faulting, disrupting the original stratigraphic position of the various lithounits (GSI 2004).

Since a mountainous terrain has been chosen as the study area, ambiguities related to drainage extraction are expected to be absent. In a flat terrain, the drainage networks derived usually show wide deviations from reality (Rahman et al. 2010).

These problems may be minimized by using a dataset of continuously distributed elevation data across an area. Moreover, the possible errors that may occur due to inaccurate channel network demarcation, masking effects of vegetation or cultivation and cartographic compulsions in map preparation, may be minimized by employing such continuous elevation data. Modern day DEMs have come as an answer to these issues.

This analysis of DEM derived information in the present study is thus topical and of importance as it influences analysis of landscape configuration. This paper provides a comparative study of different available or derived DEMs (from SRTM, ASTER, Cartosat-1 tiles and SoI, USAMS topographical maps), through extraction of stream networks and terrain aspects, enumeration of different morphometric indices, and their eventual comparison.

Processing and ensuing analysis of the above datasets has been performed sequentially using the following methods.

DEM processing and extraction of drainage networks

A flowchart schematically shows the methodology followed  for the extraction of drainage networks and surface attributes from DEMs in a GIS environment (Fig. 2). The SRTM DEM of the study area is first preprocessed through the operations of filling the data gaps, pit removal–depression filling, and finding outlet cells in an iterative manner (O’Callaghan and Mark 1984; Jenson and Domingue 1988). Pit removal and depression filling is a method of filtering the digital elevation data. This is done to overcome any data voids that may be present in the DEM tile and to also ensure proper channel network connectivity. Sometimes, there are some pixels in the continuous array of digital data where the value of the pixel is abnormally low or high in comparison to other neighbouring cells. These are known as data sinks or spikes respectively and these are inherent in any DEM. These need to be removed before carrying out any sort of analysis in the data (Wood 1996). Along with the SRTM data, ASTER and Cartosat-1 DEMs were also preprocessed and all the possible data sinks and spikes were removed.

The derivation of DEMs from USAMS and SoI topographical maps involved a rather time-consuming and labour-intensive technique. These maps were obtained either as a scanned raster object (in case of the USAMS Map of scale 1:250,000) or as hard-copy maps that were then scanned at 300 dots per inch (in case of the SoI maps of scale 1:50,000). These were then georeferenced using the location information (latitude and longitude) demarcated in them. The contour lines were then digitised on-screen manually in the ArcGIS environment to prepare contour maps (Fig. 3). All such vector contour datasets were converted to WGS84 datum and then processed to derive the respective surface models (Fig. 4) through triangulated interpolation and subsequent smoothening of the derived surfaces. From these surface models, the stream networks were later extracted for the respective topographical maps, as described below.

A stream network develops as an interface between the concentrative processes acting in and towards the channels, as compared to the diffusive processes acting divergently, across the surrounding hill slopes. The simplest method for specifying flow directions is to assign flow from each pixel to one of its eight neighbors, either adjacent or diagonally, in the direction of the steepest downward slope. This method, designated D8 (choosing any 1 out of 8 flow directions on the basis of the line of steepest descent), was introduced by O’Callaghan and Mark (1984) and has been widely used as it generates channel networks with no divergence, allowing water to be routed unambiguously (Band 1986). In the context of a grid, the upslope area (A) contributing to each pixel is estimated as the product of the number of pixels draining through each pixel and the pixel area. The specific catchment area (SCA) is then estimated as A/L, taking L as the pixel width (Lindsay 2009). A pointer data layer is then created that stores the flow direction of each cell in a raster grid and the topology of the flow network is thus generated (Patel and Sarkar 2009). Within the GIS environment, algorithms for flow accumulation, flow routing and flow direction analysis were run which helped to extract the drainage network. This drainage network was then ordered using the Strahler (1954) scheme of stream ordering, wherein each of the finger-tip tributaries were designated as Order 1. Where two streams of the same order meet, the resulting stream order of the subsequent unified stream increases by one. This scheme was followed to categorise the streams derived from each of the DEMs.

It is pertinent to mention here that a portion of the studied basin remains under snow cover perennially. Therefore, stream network generation from the entire DEM creates streams over areas covered by glaciers, mostly by taking potential flow-lines along either edge of the flat ice-filled valley floor or through the base of the cliffs on either valley side, where they abut onto the edge of the glacier. This causes flow-lines to be shown where streams are placed parallel to each other till they join at the glacial snout to become a single stream segment. Such a derived network is erroneous as it greatly increases the network extent spuriously and also causes inaccuracies in stream ordering and subsequent evaluation of morphometric parameters. Therefore, an ice cover mask has been used initially, to allow stream extraction only in the ice free region and a more realistic drainage network is obtained for each of the analyzed datasets. This mask was demarcated via on-screen digitisation of the snow cover, as visually interpreted from an IRS P6 LISS-III scene of October 2008, imaged over the study area, and reconfirmed from the snow cover and glacial portions demarcated in the SoI topographical sheets. This digitised polygon layer was overlain on each of the Basin DEM surfaces, to mask out the portion covered by ice, in order that spurious and parallel channel networks were not derived over and along each side of the glacial snouts, which would then skew stream ordering and network length enumerations. Furthermore, to show the amount of error that non-usage of the ice-mask creates in extraction of the drainage network and its allied morphometric parameters, a comparison has also been laid out between the extracted flowlines from the overall area (i.e., obtained without using an ice-mask and thus erroneous in overestimating drainage line number and lengths) and from just the ice-free area (more realistic and accurate).

Figures 5, 6 and Fig. 7 represent the drainage network of the Supin–Upper Tons Watershed, as extracted from SRTM, ASTER and Cartosat-1 DEM respectively—with and without the ice-cover mask and also for the resampled DEMs. Since the SRTM-DEM data has a resolution of 90 m and the ASTER and Cartosat-1 datasets are of 30 m resolution, the ASTER and Cartosat-1 datasets were resampled down to 90 m, to remove any bias in network derivation and subsequent computations in channel parameters that may occur due to this variation in resolution. Stream networks have been again subsequently re-derived for comparison from these resampled DEMs (Figs. 6c, 7c respectively).

The next exercise was to extract the stream network from the digitized contours of the USAMS and SoI topographical maps. The contours and all spot elevations were digitized, their respective elevation values input and these were then converted into a DEM via triangulated interpolation, with the resultant data being processed to extract the drainage networks as was done from the satellite-derived DEMs before (Figs. 8, 9). All the above datasets were then brought into a common reference framework (projection and datum-wise) for comparison. In total thus, there are nine separate datasets (3 downloaded DEMs, 2 resampled DEMs from finer data, 2 DEMs prepared via contour digitisation from topographical maps and 2 surveyed topographical maps), from which the various terrain and drainage attributes are subsequently derived and mapped for comparison. The notations used to refer to these datasets in this paper are as follows: A30 (ASTER 30 m DEM), A90 (ASTER resampled 90 m DEM), C30 (CartoDEM 30 m DEM), C90 (CartoDEM resampled 90 m DEM), S90 (SRTM 90 m DEM), T50A (actual digitised contour and stream network database from 1:50,000 scale topographical maps), T50D (DEM generated from digitised contours of 1:50,000 scale topographical maps, resampled to 90 m), T250A (actual digitised contour and stream network database from 1:250,000 scale topographical maps), and T250D (DEM generated from digitised contours of 1:250,000 scale topographical maps, resampled to 90 m). The basin outlines for each dataset (as presented within the foregoing figures), were also extracted in a GIS environment after demarcation and ordering of the channel network using the pour point function, wherein all the cells which drain into a particular outlet are grouped together and their combined perimeter forms the basin boundary. The areas of the basin polygons derived from each dataset thus, were then computed.

Digital Elevation Models (DEMs) have been a subject of increasing attention and utilization in the last few decades because of the relative ease in delineation, extraction and calculation of various drainage and terrain morphometric parameters from them. Keeping this fact in mind, the present study was carried out in order to find the best possible DEM for computing the morphometric attributes of drainage basins from, especially in terrains that are difficult to survey or access. After analyzing the different morphometric parameters derived from these DEMs, it can be said that the DEMs derived from the 1:50,000 topographical map and ASTER GDEM datasets are relatively more accurate and consistent. They also exhibit a certain degree of proximity to the surveyed topographical map data. If 1:50,000 scale topographical maps of an area are not available, then the ASTER GDEM 30 m followed by the 4th generation SRTM DEM 90 m provides viable alternatives to analyse the terrain attributes of the area. While India’s indigenous and freely available Cartosat-1 DEM 30 m is unable to match the accuracy and consistency of the results produced by ASTER GDEM 30 m and SRTM DEM 90 m for this study area, the difference or deficiency is however lesser than those for resampled DEMs or DEMs prepared from smaller scale 1:250,000 scale topographical maps. The 30 m ASTER DEM also proves to be viable in examining terrains at even larger scales of 1:25,000; since topographical maps at this scale are rarely available for this country, due to an incomplete coverage. For large areas, where a greater numbers of maps are involved, these DEM datasets provide a relatively quicker pathway to topographic and drainage analysis.

DEM usage always comes with some caveats however. Sharma et al. (2009) while working on contour interpolated DEMs, postulated that grid size plays an important role in measuring the vertical accuracy of the DEMs. Furthermore, the generation of DEMs from topographical sheets can induce errors or omissions in scanning, georeferencing and digitisation, all of which may affect the resultant output DEM quality and the stream network information derived from it. This is corroborated by Ahmed et al. (2010) while working on the Bandihole Sub-watershed in Karnataka, India. However, in the present study, although the SRTM and ASTER datasets provide substantial results, the Cartosat-1 dataset does not provide similarly reliable information. Inherent limitations of the Cartosat-1 dataset might have played a part in reducing the accuracy of the first generation of this Cartosat-1 DEM.